Zarr

Zarr stores multidimensional data as chunked arrays. A product can contain one or more groups, and each group can contain arrays with their own dimensions, data types, chunk shapes, and attributes.

Chunking lets readers retrieve part of a large product without reading every value. The chunk layout also affects write safety and performance because a physical chunk is the smallest shared storage unit that writers can touch.

A Firecube Zarr product keeps array data in Zarr groups and physical chunks. ChunkManager keeps run, span, and claim records beside the store.

Zarr Chunks And ChunkManager Records

A Zarr chunk is a physical block of array data. ChunkManager records are Firecube control-plane metadata stored beside the product under .firecube/. They describe runs, written spans, and active claims; they are not Zarr chunks.

This distinction matters when inspecting or recovering a product. Zarr tools read the arrays. Firecube's chunks commands inspect and manage the associated run and coordination state.

Choose A Zarr Write Model

Firecube supports two Zarr authoring classes. Parallel writes are an optional extension of one of them, not a third plugin class.

Data your plugin can supply	Write model	Write consequence
Complete xarray.Dataset batches, already ordered along the append dimension	GenericZarrIngestor (Append)	Firecube finds the end of the group and serializes dataset construction and append mutations.
A declared array schema and exact indexed write locations	DirectZarrIngestor (Region)	The plugin controls placement; serial stores may grow as later indexes are written.
The direct-write contract plus a fixed extent and deterministic indexes	Parallel Zarr Writes	Separate processes can own disjoint, chunk-aligned ranges of one group.

Choose sequential appends when complete dataset batches represent the product naturally. Choose direct writes when the plugin must control individual array writes or must remove the serialized same-group append path. Add slot-range parallelism only when several processes must write one Zarr group and the product can satisfy its fixed-layout safety requirements. A DirectZarrIngestor without the slot contract remains a serial writer.

Next Steps

• GenericZarrIngestor (Append) — understand the GenericZarrIngestor write model
• DirectZarrIngestor (Region) — understand the DirectZarrIngestor write model
• Plugin Development — choose and implement a plugin authoring class
• Performance Tuning — tune chunking, sharding, compression, and batch size